library(tidyverse)
── Attaching packages ───────────────────────────────────────────────────── tidyverse 1.3.0 ──
✓ ggplot2 3.3.0     ✓ purrr   0.3.3
✓ tibble  3.0.0     ✓ dplyr   0.8.5
✓ tidyr   1.0.2     ✓ stringr 1.4.0
✓ readr   1.3.1     ✓ forcats 0.4.0
── Conflicts ──────────────────────────────────────────────────────── tidyverse_conflicts() ──
x dplyr::filter() masks stats::filter()
x dplyr::lag()    masks stats::lag()

This notebook contains solutions to all the exercises in Thomas Lin Pederson’s online workshop

Datasets

We will use an assortment of datasets throughout the document. The purpose is mostly to showcase different plots, and less on getting some divine insight into the world. While not necessary we will call data(<dataset>) before using a new dataset to indicate the introduction of a new dataset.

Introduction

We will look at the basic ggplot2 use using the faithful dataset, giving information on the eruption pattern of the Old Faithful geyser in Yellowstone National Park.

library(ggplot2)
data("faithful")

# Basic scatterplot

ggplot(data = faithful, mapping = aes(x = eruptions, y = waiting)) + 
  geom_point()

# Data and mapping can be given both as global (in ggplot()) or per layer

ggplot() + 
  geom_point(mapping = aes(x = eruptions, y = waiting), data = faithful)

If an aesthetic is linked to data it is put into aes()

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting, colour = eruptions < 3))

If you simple want to set it to a value, put it outside of aes()

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting), color = 'steelblue1')

Some geoms only need a single mapping and will calculate the rest for you

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions))
`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

All geoms are drawn in the order they are added. The point layer is thus drawn on top of the density contours in the example below.

ggplot(faithful, aes(x = eruptions, y = waiting)) + 
  geom_density_2d() + 
  geom_point()

Exercises

Modify the code below to make the points larger squares and slightly transparent.

See ?geom_point for more information on the point layer.

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting))

Hint 1: transparency is controlled with alpha, and shape with shape

Hint 2: remember the difference between mapping and setting aesthetics

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting), shape = 15, alpha = 0.5, size = 5)

You can see the most common shapes here:

Type ?pch in the console to see more options. For example you can specify ASCII characters using the numbers 33 to 127 or typing each character in between quotation marks.

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting), shape = 64, alpha = 0.5, size = 5)

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting), shape = "@", alpha = 0.5, size = 5)

Finally, note that shapes between 21 and 25 can be filled with different colors!

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting), shape = 21, size = 10, 
             fill = "skyblue", color = "#FFFF3E",
             stroke = 2) ## controls "point" border thickness

Colour the two distributions in the histogram with different colours

Hint 1: For polygons you can map two different colour-like aesthetics: colour (the colour of the stroke) and fill (the fill colour)

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions, fill = eruptions >= 3.1), show.legend = FALSE,
                 color = "white")
`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Colour the distributions in the histogram by whether waiting is above or below 60. What happens?

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions, fill = waiting > 60))
`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Change the plot above by setting position = 'dodge' in geom_histogram() (while keeping the colouring by waiting). What do position control?

All layers have a position adjustment that resolves overlapping geoms. See more here

ggplot(faithful) + 
  geom_histogram(aes(x = eruptions, fill = waiting > 60), position = "dodge")
`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Add a line that separates the two point distributions. See ?geom_abline for how to draw straight lines from a slope and intercept.

ggplot(faithful) + 
  geom_point(aes(x = eruptions, y = waiting)) + 
  geom_abline(intercept = 125, slope = -80/4)

Stat

We will use the mpg dataset giving information about fuel economy on different car models.

Every geom has a stat. This is why new data (count) can appear when using geom_bar().

data("mpg")

ggplot(mpg) + 
  geom_bar(aes(x = class))

The stat can be overwritten. If we have precomputed count we don’t want any additional computations to perform and we use the identity stat to leave the data alone

library(dplyr)

mpg_counted <- mpg %>% 
  count(class, name = 'count')

ggplot(mpg_counted) + 
  geom_bar(aes(x = class, y = count), stat = 'identity')

Most obvious geom + stat combinations have a dedicated geom constructor. The one above is available directly as geom_col()

ggplot(mpg_counted) + 
  geom_col(aes(x = class, y = count))

Values calculated by the stat is available with the after_stat() function inside aes(). You can do all sorts of computations inside that.

ggplot(mpg) + 
  geom_bar(aes(x = class, y = after_stat(100 * count / sum(count)))) + ## WTF!
  labs(y = "%")

Many stats provide multiple variations of the same calculation, and provides a default (here, density)

ggplot(mpg) + 
  geom_density(aes(x = hwy))

While the others must be used with the after_stat() function

ggplot(mpg) + 
  geom_density(aes(x = hwy, y = after_stat(scaled)))

Exercises

While most people use geom_*() when adding layers, it is just as valid to add a stat_*() with an attached geom.

Look at geom_bar() and figure out which stat it uses as default. Then modify the code to use the stat directly instead (i.e. adding stat_*() instead of geom_bar())

ggplot(mpg) + 
  stat_count(aes(x = class))

Use stat_summary() to add a red dot at the mean hwy for each group

Hint: You will need to change the default geom of stat_summary()

ggplot(mpg) + 
  geom_jitter(aes(x = class, y = hwy), width = 0.2)
ggplot(mpg, aes(x = class, y = hwy)) + 
  geom_jitter(width = 0.2) + 
  stat_summary(fun = mean, geom = "point", color = "red", size = 3)

Scales

Scales define how the mapping you specify inside aes() should happen. All mappings have an associated scale even if not specified.

You can take control by adding one explicitly. All scales follow the same naming conventions.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class)) + 
  scale_colour_brewer(type = 'qual')  ## !

Positional mappings (x and y) also have associated scales.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  scale_x_continuous(breaks = c(3, 5, 6)) + 
  scale_y_continuous(trans = "log10")

Exercises

Use RColorBrewer::display.brewer.all() to see all the different palettes from Color Brewer and pick your favourite. Modify the code below to use it

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, color = class)) + 
  scale_colour_brewer(type = "qual", palette = "Spectral")

Modify the code below to create a bubble chart (scatterplot with size mapped to a continuous variable) showing cyl with size. Make sure that only the present amount of cylinders (4, 5, 6, and 8) are present in the legend.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class)) + 
  scale_colour_brewer(type = 'qual')

Hint: The breaks argument in the scale is used to control which values are present in the legend.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class, size = cyl)) + 
  scale_colour_brewer(type = 'qual') + 
  scale_size(breaks = c(4, 5, 6, 8))

Note that the increase from 4 to 5 looks bigger than the increase from 5 to 6 (it’s not). However, it does look weird. We can fix this by giving “0” an area of zero.

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class, size = cyl)) + 
  scale_colour_brewer(type = 'qual') +
  scale_size_area(breaks = c(4, 5, 6, 8))  ## ensures 0 is mapped to 0 size

Explore the different types of size scales available in ggplot2. Is the default the most appropriate here?

  • scale_size()

  • scale_radius() (area size increases exponentially, not recommended)

  • scale_size_binned()

  • scale_size_area()

  • scale_size_binned_area()

Modify the code below so that colour is no longer mapped to the discrete class variable, but to the continuous cty variable. What happens to the guide?

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = class, size = cty))
ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, color = class, size = cty))

We have a gradient color legend and a size legend with breaks 10, 15, 20, 25, etc.

The type of guide can be controlled with the guide argument in the scale, or with the guides() function. Continuous colours have a gradient colour bar by default, but setting it to legend will turn it back to the standard look. What happens when multiple aesthetics are mapped to the same variable and uses the guide type?

The guides are integrated!

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = cty, size = cty)) + 
  scale_color_gradient(guide = "legend")

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy, colour = cty, size = cty)) + 
  guides(color = "legend")

Facets

The facet defines how data is split among panels. The default facet (facet_null()) puts all the data in a single panel, while facet_wrap() and facet_grid() allows you to specify different types of small multiples

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_wrap(~ class)

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_grid(year ~ drv)

Exercises

One of the great things about facets is that they share the axes between the different panels. Sometimes this is undesirable though, and the behaviour can be changed with the scales argument.

Usually the space occupied by each panel is equal. This can create problems when different scales are used. Modify the code below so that the y scale differs between the panels in the plot. What happens?

ggplot(mpg) + 
  geom_bar(aes(y = manufacturer)) + 
  facet_grid(class ~ .)
ggplot(mpg) + 
  geom_bar(aes(y = manufacturer)) +
  facet_grid(class ~ ., scales = "free_y")

Use the space argument in facet_grid() to change the plot above so each bar has the same width again.

ggplot(mpg) + 
  geom_bar(aes(y = manufacturer)) +
  facet_grid(class ~ ., scales = "free_y", space = "free") +
  theme(strip.text.y = element_text(angle = 0))  ## bonus

Facets can be based on multiple variables by adding them together. Try to recreate the same panels present in the plot below by using facet_wrap()

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_grid(year ~ drv)

ggplot(mpg) + 
  geom_point(aes(x = displ, y = hwy)) + 
  facet_wrap(~ year + drv)

Coordinates

The coordinate system is the fabric you draw your layers on in the end. The default coord_cartesion provides the standard rectangular x-y coordinate system.

Changing the coordinate system can have dramatic effects.

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  coord_polar()

Polar coordinates interpret x and y as radius and angle.

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  coord_polar(theta = 'y') + 
  expand_limits(y = 70)

You can zoom both on the scale…

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  scale_y_continuous(limits = c(0, 40))
Warning: Removed 3 rows containing missing values (geom_bar).

and in the coord. You usually want the latter as it avoids changing the plotted data

ggplot(mpg) + 
  geom_bar(aes(x = class)) + 
  coord_cartesian(ylim = c(0, 40))

This happens because scales transform data at the begining, whereas coordinates transform data at the end.

Exercises

In the same way as limits can be set in both the positional scale and the coord, so can transformations, using coord_trans(). Modify the code below to apply a log transformation to the y axis; first using scale_y_continuous(), and then using coord_trans(). Compare the results — how do they differ?

ggplot(mpg) + 
  geom_point(aes(x = hwy, y = displ))
ggplot(mpg) + 
  geom_point(aes(x = hwy, y = displ)) + 
  scale_y_continuous(trans = "log")

ggplot(mpg) + 
  geom_point(aes(x = hwy, y = displ)) + 
  coord_trans(y = "log")  ## this one looks better

Coordinate systems are particularly important in cartography. While we will not spend a lot of time with it in this workshop, spatial plotting is well supported in ggplot2 with geom_sf() and coord_sf() (which interfaces with the sf package). The code below produces a world map. Try changing the crs argument in coord_sf() to be '+proj=robin' (This means using the Robinson projection).

data("world", package = "spData")

ggplot(world) + 
  geom_sf() 

ggplot(world) + 
  geom_sf() +
  coord_sf(crs = "+proj=robin")

ggplot(world) + 
  geom_sf() +
  coord_sf(crs = "+proj=laea +x_0=0 +y_0=0 +lon_0=-74 +lat_0=40")

Maps are a huge area in data visualisation and simply too big to cover in this workshop. If you want to explore further I advice you to explore the r-spatial wbsite as well as the website for the sf package

Theme

Theming defines the feel and look of your final visualisation and is something you will normally defer to the final polishing of the plot. It is very easy to change looks with a prebuild theme,

ggplot(mpg) + 
  geom_bar(aes(y = class)) + 
  facet_wrap(~year) + 
  theme_minimal()

Further adjustments can be done in the end to get exactly the look you want

ggplot(mpg) + 
  geom_bar(aes(y = class)) + 
  facet_wrap(~year) + 
  labs(title = "Number of car models per class",
       caption = "source: http://fueleconomy.gov",
       x = NULL,
       y = NULL) +
  scale_x_continuous(expand = c(0, NA)) + 
  theme_minimal() + 
  theme(
    text = element_text('Avenir Next Condensed'),
    strip.text = element_text(face = 'bold', hjust = 0),
    plot.caption = element_text(face = 'italic'),
    panel.grid.major = element_line('white', size = 0.5),
    panel.grid.minor = element_blank(),
    panel.grid.major.y = element_blank(),
    panel.ontop = TRUE
  )

Exercises

Themes can be overwhelming, especially as you often try to optimise for beauty while you learn. To remove the last part of the equation, the exercise is to take the plot given below and make it as hideous as possible using the theme function. Go absolutely crazy, but take note of the effect as you change different settings.

g <- ggplot(mpg) + 
  geom_bar(aes(y = class, fill = drv)) + 
  facet_wrap(~year) + 
  labs(title = "Number of car models per class",
       caption = "source: http://fueleconomy.gov",
       x = 'Number of cars',
       y = NULL)

g + theme(
  text = element_text("Comic Sans MS", color = "orange"),
  axis.text = element_text(color = "white"),
  panel.background = element_rect("yellow"),
  legend.background = element_rect(fill = "yellow", linetype = "dotted", 
                                   color = "white", size = 2),
  strip.text = element_text(face = 'bold', hjust = 1, angle = 20),
  plot.background = element_rect("black")
  )